search

apl-ocean-engineering / lsd-slam

My experiments with LSD-SLAM
GNU General Public License v3.0
25 stars 5 forks source link

Infinite loop on LSD_Machine data set #3

Open apatzer opened 8 years ago

apatzer commented 8 years ago

Running LSD_SLAM with the machine data set from @JakobEngel, on frame 141-142, the system gets stuck in a loop forever:

2016/05/09 16:51:00 125522296   DEBUG [OptimizationThread.cpp->callbackIdle:38] "Running short optimization"
2016/05/09 16:51:00 125522671   DEBUG [OptimizationThread.cpp->optimizationIteration:123]   "did 1 optimization iterations. Max Pose Parameter Change: 0.027259; avgChange: 0.005622. Waiting for addition to graph.
"
2016/05/09 16:51:02 127525521   DEBUG [OptimizationThread.cpp->callbackIdle:38] "Running short optimization"
2016/05/09 16:51:02 127525900   DEBUG [OptimizationThread.cpp->optimizationIteration:123]   "did 1 optimization iterations. Max Pose Parameter Change: 0.027259; avgChange: 0.005622. Waiting for addition to graph.

It seems stuck in optimization without being satisfied. Suggestions?

amarburg commented 8 years ago

No immediate suggestions, I've been using my data sets instead of "Machine". I'm curious if I can replicate it on my Linux box.

Doing side-by-side testing, it's clear the refactor wasn't perfect --- I'm always looking for places where the refactored code has regressed from the original. Changing the synchronization between the threads means it will never run exactly as the original ran, but it should have comparable performance (or maybe I should say .... I'm eager for someone to show me how/why/where the refactored version has worse performance).